Nandu

Agent infrastructure.Licensed and implemented.

Six building blocks that turn agent work from a personal speed-up into something a whole team compounds on. This is the stack we run our own company on.

The stack

Built for ourselves first. Licensed now.

None of this started as a product. Each block exists because a small team wanted to ship at the pace of several and kept hitting the same walls. The blocks run in our production every day, which is also why we can set them up in your environment instead of handing over an archive and wishing you luck.

The building blocks

01

Nandu Development Framework

The methodology, as installable software

Spec-driven and test-driven development with agents, packaged: agent definitions, commands, project conventions, and a CLI that installs and updates the whole set. Your team pins a version and updates on its own schedule.

How you get it

The versioned framework, the CLI with its migration system, read-only access to the framework repository, a field-notes repo that feeds your observations back into the framework, and onboarding through the first project.

Learn More
02

Knowledge Infrastructure

Shared long-term memory for the team and its agents

The layer every session loads at the start: decisions, conventions, client and product knowledge. A strict purpose filter keeps it a context layer people trust instead of a wiki nobody opens.

How you get it

The structure, the curation rules that keep it useful, the automation that keeps it current, and setup with your team's real knowledge.

03

Capability Distribution

Write a capability once, every engineer gets it

Skills and procedures are written once, versioned centrally, and distributed to every engineer's environment. Who receives what is controlled by role and access, so each person gets exactly the capabilities they are entitled to.

How you get it

The distribution pipeline, versioning and registry discipline, the entitlement configuration, and packaging for your team's environments.

04

Workbench

One board for every improvement in flight

The state layer for improvement work: what you are changing, what you tried, what came of it. Structured so an agent can pick up an item and run with it while people keep the final gate.

How you get it

The board and dossier conventions, the templates, and the working discipline that keeps changes from happening outside the board.

05

Flock

Work automation that drives sessions for you

The orchestration layer that starts, briefs, verifies, and closes agent working sessions, so your people spend attention only where a human decision is genuinely needed. What it cannot answer it escalates, and it learns the answer for the next time.

How you get it

The orchestration system, the manager instruction sets, the verification checks, and setup in your environment.

06

Dreamer

Every session reviewed, every lesson kept

Dreamer reads each agent session after it ends and asks what should improve: a convention that caused friction, a missing check, a skill worth writing. Its findings land in the Workbench as improvement items. Each session then pays twice, once in the work itself and once in what your setup learns from it. The learnings stay in your infrastructure and are never shared with us.

How you get it

The session review pipeline, the distillation instructions, the Workbench wiring, and configuration for your environment.

How you get it

Licensed and implemented, never handed over cold

The blocks differ in shape. The development framework installs through its own CLI. The knowledge layer and the Workbench are as much practice as software. Orchestration and Dreamer need real setup in your environment. Every engagement therefore includes implementation, because infrastructure without the practice around it turns into shelfware within a quarter.

Common questions

Does this work with the AI tools we already use?

Yes. Knowledge, conventions, specs, and orchestration state live as plain files, not inside any one vendor's tooling. During setup we map everything onto whatever your team runs today.

Do we have to take all six blocks?

No. Most teams start with the development framework and the knowledge layer, because those change the daily work fastest. Orchestration and Dreamer usually follow once the practice underneath is solid.

How is it licensed?

Licence plus implementation, scoped per engagement. It depends on team size, the blocks you take, and how deep the transition goes. There is no per-seat SaaS tier. Most of the value transfer is practice, and practice does not price by seat.

What if our codebase looks nothing like yours?

That is the normal case. The framework itself contains nothing specific to us or to any client. What belongs to your project lives in your project's files. That separation is what makes the whole thing portable.

Who is this for?

Engineering teams past the pilot stage, where individual speed-ups did not turn into team throughput. It pays off most where several people share one codebase and consistency matters more than raw speed.

See it running

We will walk your team through the stack as we use it ourselves, and figure out which blocks move your throughput first.